NVIDIA diffusion language model Nemotron TwoTower achieves 2.42x LLM inference throughput without a full retraining run, ...

DSpark can make decoding faster, but acceptance quality still determines how much speed the system actually realizes.

DeepSeek speculative decoding framework DSpark went live June 27 on V4-Flash and V4-Pro, reporting up to 85 percent faster ...

LongCat 2.0 is Meituan's 1.6T-parameter MoE model with a native 1M-token context, trained entirely on domestic AI ASIC ...

Local AI inference at 32B-parameter quality, no cloud API required: University of Waterloo researchers released PAW on July 2 ...

This is the inside story of how a group of academic visionaries, engineers, and industry veterans are building the foundational infrastructure for India's own "DeepSeek moment".

Chinese AI lab Zhipu AI releases GLM-5.2 with a stable 1-million-token context under the MIT license. On hours-long coding tasks, the open-source model trails Anthropic's Opus models by just a few ...

Overview of the paper: What exactly is Fugu? Currently, in the AI industry, companies like OpenAI (GPT series), Anthropic (Claude series), and Google (Gemini series) are developing massive AI models ...

MiniMax M3, a new multimodal model developed by MiniMax, is now available on NVIDIA’s accelerated infrastructure and supports advanced processing of text, images, and video. With 428 billion ...

AI can help teams move faster, but speed alone does not solve the harder problem of understanding what has actually been ...